In traditional circuit switched telephony, a continuous data “pipe” is provided through the Public Switched Telephone Network (PSTN) to guarantee the flow of the PCM voice data. Internet telephony on the other hand must overcome a variety of impairments to the regular and timely delivery of voice data packets to the far end. These impairments are inherent in current Internet architecture, which provides a best-effort delivery service without any guarantees regarding the delivery of voice packets. Additionally, the transport of the voice packets is constrained by the amount of bandwidth available in the network connection, the delay that the packet experiences and any packet loss or corruption that occurs. In general, the measure of the quality of a data network to transport voice data packets quickly and consistently is referred to as the network's quality of Service (QoS).
A variety of network conditions affect the QoS of a connection. The bandwidth (BW) is the measure of the number of bits per second that can flow through a network link at a given time. Available bandwidth is limited by both the inherent capacity of the underlying network as well as other traffic along that route. End-to-End bandwidth from sender to receiver (the “call path”) will be determined by the slowest link on the entire route. For example, a dialup connection to the ISP with an ideal bandwidth of 56 kilobits per second (kb/s) may be the slowest link for a user. However, the bandwidth actually available to a VoIP application on this link at a particular time will be lower if a larger file transfer is taking place at that time.
The bandwidth usage per channel is determined primarily by the compressor/decompressor (CODEC) used to digitize and compress voice data and its associated overhead. Table 1 lists the one-way bandwidth requirements of three popular CODECs and a Mean Opinion Score (MOS) based on the ITU-T recommendation for measuring voice quality (higher MOS values indicate better quality).
TABLE 1
As illustrated in Table 1, CODECs such as G.723 and G.729 significantly reduce the data bandwidth required. There is, however a general tradeoff, between using a high compression CODEC (with its low bandwidth usage) and voice quality. The high compression CODECs typically have slightly reduced voice quality (as reflected in the MOS rating), and introduce additional delay due to the added computational effort. The highest bandwidth is required by the minimal compression G.711 CODEC, which is the standard toll quality CODEC.
Another factor in bandwidth usage is the overhead introduced by different IP layers. Most CODECs operate by collecting a block of voice samples and then compressing this block to produce a frame of coded voice. As this media frame is prepared for transport over IP, different protocol layers add their own headers to the data to be able to recreate the voice stream at the destination. FIG. 1 illustrates how an IP datagram carrying a single G.723.1 version-1 frame will look like on a dial-up line.
Protocol overhead can be reduced by including more than one media frame per datagram (or packet). This also reduces the number of packets sent per second and hence the bandwidth usage. FIG. 2 illustrates an example of how the bandwidth usage is reduced when using 2, 3 and 4 frames per IP datagram using G.723.1 v1 CODEC. This improved efficiency comes at the cost of increased delay, but also has a positive side effect of improving jitter-tolerance. The effect of delay and jitter on voice quality is described below.
Delay along the voice transmission call path can significantly affect voice quality. If the delay is too large, for example greater than 400 ms (ITU-T recommendation), interactive communication will be impossible. Many factors contribute to delay in VoIP, the most important being the delay experienced by VoIP media packets on the network. Another source of delay is the CODEC used for processing voice. High compression CODECs introduce more delay than low compression CODECs.
VoIP media packets comprising a data stream may not experience the same delay. Some packets may be delayed more than others due to instantaneous network usage and congestion or as a result of traversing different routes through the network. This variance from the average delay is called jitter. Voice CODECS will produce poor voice output if the input packet stream is not delivered at the exact play-out time. A jitter buffer at the receiver can smooth but this variation but it adds some more delay. If the jitter is larger than what the buffer can handle, the jitter buffer may underflow or overflow resulting in packet loss.
QoS is also degraded by packet loss. The most common cause of packet loss on land-based networks is the overloading of a router queue along the transmission call path. In this case the router will discard packets. On land-based networks, packet loss is therefore a sign of network congestion. Packets can also be lost because of corruption. Internet routers are programmed to discard corrupted packets. Voice CODECS can generally cope with small random packet losses, by interpolating the lost data. Large packet loss ratio or burst packet loss can severely degrade voice quality. The exact limits vary by the CODEC used but generally, low compression CODECS are more tolerant to packet loss.
The lack of QoS guarantees on the Internet has been a major challenge in developing VoIP applications. IETF is working on a number of proposals to help guarantee the quality of service that time critical data such as VoIP services require, including:
Differentiated Service (“Diffserv”) which instructs the network routers to route based on priority bits in the packet header.
Integrated Services and RSVP to set up end-to-end virtual channels that have reserved bandwidth similar to circuit-switched telephony
Multi-protocol Label switching which uses labels inserted into the packets to route traffic in an efficient way
These services are, however, not currently available on the present day Internet. VoIP applications on end systems are required to work around the hurdles presented to regular and timely data flow. The Internet offers a best effort delivery service. So long as sufficient bandwidth is available, VoIP traffic can flow smoothly with an acceptable QoS. If the bandwidth is constrained, the effects described above will result in degraded voice quality.
What would be desirable are means to allow VoIP applications to sense the current call path bandwidth and to adapt in real-time the transmission rate to utilize that bandwidth.